Throughout this project, I have been saying that the oscillator I am using, the DS32kHz, is accurate to 7.5 parts per million, or 4 minutes per year. Having run the clock continuously for about 3 weeks now I would expect the clock to have drifted by approximately 14 seconds. However, measuring the clock against the clock on my phone I have found that it has drifted by approximately only two seconds.
My first explanation for this was that I must have gotten lucky and had bought a DS32kHz that was well within spec but digging further into the datasheet I found this:
Here you can see that, although the worst case tolerance for the device is 7.5ppm from -40°C to 0°C and from +40°C to +85°C, between 0°C and +40°C, the device actually has a tolerance of only 2ppm or 1 minute per year. The temperate in my room is obviously always in that range and therefore, over three weeks, the expected deviation is about 4 seconds. While my deviation of only 2 seconds is still well within spec, it is not dissimilar to the expected deviation.